STATISTICAL MODELS OF AT-GRADE INTERSECTION ACCIDENTS

The objective of this research was to develop statistical models of the relationship between traffic accidents and highway geometric elements for at-grade intersections. These models also incorporated the effect of traffic control features and traffic volumes on intersection accidents. The data base used to develop the models was obtained from the California Department of Transportation. Field data were also collected for a sample of urban, four-leg, signalized intersections to provide data on additional geometric design variables and turning-movement counts that were not available from existing highway agency files. The statistical modeling approaches used in the research included Poisson, lognormal, negative binomial, and logistic regression, as well as discriminant and cluster analysis. Regression models of the relationships between accidents and intersection geometric design, traffic control, and traffic volume variables were found to explain between 16 and 38% of the variability in the accident data. However, most of that variability was explained by the traffic volume variables considered; geometric design variables accounted for only a very small additional portion of the variability. An evaluation of hard-copy police accident reports by three independent reviewers for a sample of eight urban, four-leg, signalized intersections found that only 5 to 14% of the accidents had causes that appeared to be related to geometric design features of the intersections.